https://umd.zoom.us/j/
In the evolving landscape of artificial intelligence (AI), this doctoral thesis presents a comprehensive exploration of generative AI, focusing on the enhancement of security through innovative watermarking techniques, the expansion of algorithmic capabilities, and the pioneering of new approaches in creative image generation. It introduces certified watermarking techniques to safeguard Deep Neural Networks (DNNs) from intellectual property violations, representing a significant advancement in the protection of AI innovations. This research further delves into the potential of neural networks to transcend traditional pattern recognition, by engaging in complex problem-solving and exhibiting advanced reasoning, thereby extending the boundaries of algorithmic synthesis.
Additionally, the thesis explores novel territories in image generation with diffusion models, employing deterministic transformations to generate images thus expanding the working mechanics behind these diffusion models. Furthermore, this research facilitates the creation of highly specific and contextually relevant imagery given any condition with minimal necessity for model retraining. This approach enhances the flexibility and applicability of diffusion models, opening new avenues for creative expression within generative AI.
Through these investigations, the thesis contributes significantly to the fields of AI watermarking, algorithm learning capabilities, and creative image generation. It offers a holistic strategy for the development of next-generation AI technologies, addressing the crucial challenges of securing and innovating within the AI domain while enhancing the creative potential of generative models. This research not only aligns with the thesis' objective to advance watermarking, algorithm synthesis, and diverse generative strategies but also lays a solid foundation for future advancements in the realm of artificial intelligence, marking a pivotal step towards realizing the full potential of generative AI.